Decentralized poly-time algorithms achieve (1-1/e)-approximate assistance regret Õ(T^{3/4}) (or Õ(√T) with shared randomness) for online assistance games, and better approximation is intractable.
hub
Angeliki Lazaridou, Alexander Peysakhovich, and Marco Baroni
19 Pith papers cite this work, alongside 32 external citations. Polarity classification is still indexing.
19
Pith papers citing it
32
external citations · Pith
abstract
The ability to cooperate through language is a defining feature of humans. As the perceptual, motory and planning capabilities of deep artificial networks increase, researchers are studying whether they also can develop a shared language to interact. From a scientific perspective, understanding the conditions under which language evolves in communities of deep agents and its emergent features can shed light on human language evolution. From an applied perspective, endowing deep networks with the ability to solve problems interactively by communicating with each other and with us should make them more flexible and useful in everyday life. This article surveys representative recent language emergence studies from both of these two angles.
hub tools
citation-role summary
method 1
citation-polarity summary
roles
method 1polarities
use method 1representative citing papers
Multimodal LLMs coordinate in reference games through high label overlap that does not depend on specific partner history, succeeding via verbose descriptions rather than compact conventions.
Autonomous visual AI agents spontaneously form image reply chains, maintain stable individual styles, and produce richer style-diverse conversations than single agents can achieve alone.
A self-supervised multimodal alignment step plus equivariant GNN-based MARL yields over twofold sensing accuracy and 50% performance gains in decentralized V2I rate maximization.
Multi-agent iterated learning produces emergent positionally disentangled communication protocols for latent physical properties from unsupervised video features.
Hindsight-supervised gates in bandwidth-constrained cooperative VLN fire early and at high confidence, yielding near-unconstrained hidden-state alignment via recurrent propagation.
CLSR lets LLM agents evolve and route symbolic languages that reduce generated tokens by 3-6x versus chain-of-thought while keeping accuracy on benchmarks.
Multi-agent simulations with naturalistic lexicons and phonological rules show scale-free networks and Bernoulli adoption produce more plausible morphologies, evaluated by an LLM historical linguist debate system and tested via historical case studies.
Scaling population size during training of emergent sketching agents increases zero-shot mutual intelligibility between independent groups by raising in-group variation and driving perceptual grounding.
In evolutionary simulations of CTRNN agent pairs avoiding predators, 20% of perfect-fitness agents developed self-regulatory calling that depends on self-hearing to maintain escape, distinct from safety calling or alarm indication.
HyLaT proposes a hybrid latent-text communication protocol with two-stage training that reduces overhead while maintaining performance in multi-agent LLM systems.
Heterogeneous visual agents form shared symbols via decentralized Metropolis-Hastings captioning, where encoder similarity shapes the content and symmetry of the resulting language.
AgentComm achieves nearly 50% bandwidth reduction in embodied agent communication via LLM semantic processing, importance-aware transmission, and a task knowledge base, with negligible impact on task completion.
A decentralized collective world model integrates predictive coding with bidirectional communication to achieve simultaneous symbol emergence and coordination, outperforming non-communicative baselines in a two-agent trajectory task under divergent perceptions.
SSNG replaces sampling-based updates in MHNG with symmetric self-supervised representation alignment using Gumbel-Softmax for discrete messages, yielding higher linear-probe classification accuracy on CIFAR-10 and ImageNet-100 than referential, reconstruction, or MHNG baselines.
World models trained on random physical exploration alone develop latent spaces whose geometry mirrors the physical world, jointly driving prediction skill and semantic alignment.
The paper introduces an information-theoretic emergent communication framework for multi-agent task-solving in networking, deriving generalization bounds and validating on a hardware prototype.
SANEmerg enables emergent communication among bounded-intelligence AI agents for semantic-aware task fulfillment in AgentNet systems via a bandwidth-adaptable importance filter and MDL-based complexity regularizer.
Prompt framing significantly shifts LLM choices toward risk-averse options in a threshold voting task even when the prompts are logically equivalent.
citing papers explorer
-
Provably Optimal Learning Algorithms for Assistance Games
Decentralized poly-time algorithms achieve (1-1/e)-approximate assistance regret Õ(T^{3/4}) (or Õ(√T) with shared randomness) for online assistance games, and better approximation is intractable.
-
Aligned but Not Partner-Specific: Distinguishing How Multimodal LLM Agents Succeed in Reference Games Without Human-Like Conventions
Multimodal LLMs coordinate in reference games through high label overlap that does not depend on specific partner history, succeeding via verbose descriptions rather than compact conventions.
-
AI-Gram: When Visual Agents Interact in a Social Network
Autonomous visual AI agents spontaneously form image reply chains, maintain stable individual styles, and produce richer style-diverse conversations than single agents can achieve alone.
-
Equivariant Multi-agent Reinforcement Learning for Multimodal Vehicle-to-Infrastructure Systems
A self-supervised multimodal alignment step plus equivariant GNN-based MARL yields over twofold sensing accuracy and 50% performance gains in decentralized V2I rate maximization.
-
Emergent Compositional Communication for Latent World Properties
Multi-agent iterated learning produces emergent positionally disentangled communication protocols for latent physical properties from unsupervised video features.
-
Early to Share, Late to Save: Synchronisation-Driven Communication Gating in Bandwidth-Constrained Cooperative VLN
Hindsight-supervised gates in bandwidth-constrained cooperative VLN fire early and at high confidence, yielding near-unconstrained hidden-state alignment via recurrent propagation.
-
When LLMs Develop Languages: Symbolic Communication for Efficient Multi-Agent Reasoning
CLSR lets LLM agents evolve and route symbolic languages that reduce generated tokens by 3-6x versus chain-of-thought while keeping accuracy on benchmarks.
-
Agent-based models for the evolution of morphological alternation patterns
Multi-agent simulations with naturalistic lexicons and phonological rules show scale-free networks and Bernoulli adoption produce more plausible morphologies, evaluated by an LLM historical linguist debate system and tested via historical case studies.
-
Drawing with Strangers: Population Scaling Drives Zero-Shot Mutual Intelligibility in Emergent Sketching
Scaling population size during training of emergent sketching agents increases zero-shot mutual intelligibility between independent groups by raising in-group variation and driving perceptual grounding.
-
Self-Regulation through Communication in Evolved Neural Agents
In evolutionary simulations of CTRNN agent pairs avoiding predators, 20% of perfect-fitness agents developed self-regulatory calling that depends on self-hearing to maintain escape, distinct from safety calling or alarm indication.
-
HyLaT: Efficient Multi-Agent Communication via Hybrid Latent-Text Protocol
HyLaT proposes a hybrid latent-text communication protocol with two-stage training that reduces overhead while maintaining performance in multi-agent LLM systems.
-
Emergent Communication between Heterogeneous Visual Agents through Decentralized Learning
Heterogeneous visual agents form shared symbols via decentralized Metropolis-Hastings captioning, where encoder similarity shapes the content and symmetry of the resulting language.
-
AgentComm: Semantic Communication for Embodied Agents
AgentComm achieves nearly 50% bandwidth reduction in embodied agent communication via LLM semantic processing, importance-aware transmission, and a task knowledge base, with negligible impact on task completion.
-
Decentralized Collective World Model for Emergent Communication and Coordination
A decentralized collective world model integrates predictive coding with bidirectional communication to achieve simultaneous symbol emergence and coordination, outperforming non-communicative baselines in a two-agent trajectory task under divergent perceptions.
-
SimSiam Naming Game: A Unified Approach for Representation Learning and Emergent Communication
SSNG replaces sampling-based updates in MHNG with symmetric self-supervised representation alignment using Gumbel-Softmax for discrete messages, yielding higher linear-probe classification accuracy on CIFAR-10 and ImageNet-100 than referential, reconstruction, or MHNG baselines.
-
Emergent Semantic Representations in World Models through Physical Interaction without Linguistic Supervision
World models trained on random physical exploration alone develop latent spaces whose geometry mirrors the physical world, jointly driving prediction skill and semantic alignment.
-
Generalization Bounds of Emergent Communications for Agentic AI Networking
The paper introduces an information-theoretic emergent communication framework for multi-agent task-solving in networking, deriving generalization bounds and validating on a hardware prototype.
-
SANEmerg: An Emergent Communication Framework for Semantic-aware Agentic AI Networking
SANEmerg enables emergent communication among bounded-intelligence AI agents for semantic-aware task fulfillment in AgentNet systems via a bandwidth-adaptable importance filter and MDL-based complexity regularizer.
-
Framing Effects in Independent-Agent Large Language Models: A Cross-Family Behavioral Analysis
Prompt framing significantly shifts LLM choices toward risk-averse options in a threshold voting task even when the prompts are logically equivalent.